Well, now I need to update my book. To my knowledge, this is the first time Sam Altman hasn't downplayed or dismissed AI existential risk since early 2023. TBC, I think it's good of Altman to say this if that's what he actually believes...

POV: Your LLM agent is dividing a by b

Comparing Claudes: Sonnet 4.5 has a 8% longer time horizon point estimate than Opus 4.1. However, this difference isn’t statistically significant. Its point estimate is 66% longer than Sonnet 4. This is statistically significant, with 4.5 beating 4 in 99% of bootstrap samples.

We evaluated Gemini 2.5 Deep Think on FrontierMath. There is no API, so we ran it manually. The results: a new record! We also conducted a more holistic evaluation of its math capabilities. 🧵

One AI year is seven Internet ones.

This is very true. Wow. It makes me wonder whether I should leave this place.

It should not be illegal to give bad advice! That’s how you end up with a society where no-one is willing to give any advice at all.

Elon Musk Matt Walsh Grok We’re not dead yet, and I think you can do a lot to help. I think you should: 1. Continue speaking up about the threat of human extinction from ASI. 2. Call for an international treaty to ban the development of ASI. 3. Shut down xAI, to show you’re serious, and to reduce

One Tuesday night, as my wife and I sat down for dinner, a sheriff’s deputy knocked on the door to serve me a subpoena from OpenAI. I held back on talking about it because I didn't want to distract from SB 53, but Newsom just signed the bill so... here's what happened: 🧵

At what is possibly a risk to my whole career I will say: this doesn't seem great. Lately I have been describing my role as something like a "public advocate" so I'd be remiss if I didn't share some thoughts for the public on this. Some thoughts in thread...

Proud to announce that I've been awarded a CIFAR Chair in my new position at Mila/UdeM! Canada has a long track record of providing excellent funding opportunities for academic AI research; this is a big reason why Canada punches above its weight in AI.

You can survive with one kidney, so how important is the second one? Kidney donors have an increased risk of high blood pressure, proteinuria (losing proteins in urine), end-stage kidney disease, and – ultimately – shorter life expectancy The second kidney is pretty important

Objectively speaking I should prob. be honeymoon-posting about contentious topics a bit less less, and just sticking to pics of my beautiful wife ... but she's away for a minute and I'll just say one more thing on the OpenAI erotic content thing ...

😲 Whoopie Goldberg recommends “If Anyone Builds It, Everyone Dies” by Eliezer Yudkowsky ⏹️ and Nate Soares ⏹️ on The View

Thrilled to share I’ve joined OpenAI for Science, a new team building AI systems to advance scientific reasoning and accelerate discovery in math and physics. 🧵

Surprised that Microsoft will own so much of OpenAI.

Okay like... I woke up tonight, I'm pretty sure I really woke up and it wasn't just a dream, and I went to the toilet, and outside I heard some water dropping since it had rained. And the dropping happened at a fixed frequency and somehow -- no idea why -- every drop made a sound

I feel like this tweet has meme-potential.

Is this account run by Lighthaven itself, or is it some kind of satire?

New post on RL scaling: Careful analysis of OpenAI’s public benchmarks reveals RL scales far worse than inference: to match each 10x scale-up of inference compute, you need 100x the RL-training compute. The only reason it has been cost-effective is starting from a tiny base. 🧵